Course: Large Language Models and Generative AI for NLP — 2025

This year we-Aarne Talman (AMD), Jussi Karlgren (AMD), and myself (TomTom -> Aiven)-have been teaching the LLM and Gen AI course for the second year in a row, for students of different departments of the University of Helsinki.

In my 2-week part, I focused again on RAG (Retrieval Augmented Generation) and Applications. One change since 2024 was the addition of Agentic RAG, where the following basic RAG diagram:

Vanilla RAG workflow

transforms into an RAG with AI Agents, where each AI Agent has these components:

Components of an AI Agent

Comparing vanilla RAG and Agentic RAG:

How Vanilla RAG compares to Agentic RAG (courtesy: Weaviate blog)

I have also sped up the RAG UI to support the local Ollama model in addition to the Gemma-2B one.
All the code and slides are online, together with the project assignments: https://github.com/Helsinki-NLP/LLM-course-2025

Comments

[Dmitry Kan]

Btw, week 5 (RAG) and week 6 (LLM applications) can be found on Vector Podcast: youtube.com/playlist?list=PL8mdW7N...